JP2002236497A - Noise reduction system - Google Patents

Abstract

PROBLEM TO BE SOLVED: To set the best noise reduction coefficient for a vehicle of interest. SOLUTION: Noise reduction coefficients are stored in a coefficient bank 54 and noise data specifying a noise signal referred to when the noise reduction coefficients are determined are stored in a noise data bank 55 while being made to correspond to the coefficients. Pattern matching/decision parts 57 and 58 finds the noise signal which is the closest to a noise signal detected by a microphone 51 and stored in the noise data bank and a coefficient selection part 59 reads the noise reduction coefficient corresponding to the noise signal out of the coefficient bank and inputs it to a noise reduction part 52, which removes the noise signal from a detected sound signal according to the inputted noise reduction coefficient.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a noise reduction system, and more particularly to a noise reduction system for removing a noise signal from a detected sound signal detected in a passenger compartment.

[0002]

2. Description of the Related Art Speech recognition devices have become popular as human interfaces (input / output means) for in-vehicle devices such as navigation systems. The recognition performance of the speech recognition device is reduced due to disturbances other than the speech to be recognized. For this reason, various noise countermeasures have been realized. For example, a technology (car noise reduction system: CNR system) for removing vehicle running noise or engine noise that adversely affects voice recognition performance from a microphone reception signal by filtering has been realized.

FIG. 3 is a configuration diagram of a conventional noise reduction device. In the figure, 1 is a microphone, 2 is a microphone amplifier, 21 is an FIR digital filter capable of setting an arbitrary filter coefficient, 22 is a coefficient database storing filter coefficients corresponding to each of a number of noise reduction characteristics, and 23 is a predetermined database. Is a control unit (CPU) for setting the filter coefficients of the digital filter 21.

Since vehicle interior noise varies depending on the type of vehicle, noise reduction characteristics for removing the noise can be classified into several categories according to the displacement of the vehicle, the shape of the vehicle body, and the like. Therefore, an optimum filter coefficient (noise reduction coefficient) for each category is obtained in advance and set in the coefficient database 22 by using the system for obtaining the noise reduction characteristic shown in FIG. Then, a noise reduction coefficient corresponding to the vehicle is selected from the coefficient database 22 and the coefficient is set in the digital filter 21. In such a state, the digital filter 21 applies a noise reduction characteristic according to the coefficient set to the detected sound signal of the microphone 1, and outputs a sound signal from which the noise signal has been removed. As a result, the SN ratio is improved and the speech recognition rate is improved.

FIG. 4 is a block diagram of a system for obtaining a filter characteristic of a noise reduction device. In the figure, 1 is a microphone, 2 is an amplifier, 3 is a signal generator (S) that generates random noise such as white noise.
G), 4 is a variable gain amplifier, 5 is a transmission circuit that simulates the propagation characteristic CS from the speaker's mouth to the microphone and gives the propagation characteristic to the target signal, and 6 is a microphone output signal that is output from the transmission circuit. Is an addition unit for adding the value to Reference numeral 7 denotes an adaptive signal processing unit, which includes an LMS operation unit 7a and an adaptive filter 7b having an FIR digital filter configuration. At the time of learning, the signal processing unit 7 converts the output signal of the addition unit 6 into a reference signal x.
The adaptive filter coefficient W is updated so that the power of the error signal e (n) is minimized as (n).

Reference numeral 8 denotes a delay processing unit to which a random noise signal output from the signal generator 3 is input. Assuming that a signal delay time which is a half of the tap length of the adaptive filter 7b is d, the delay processing unit 8 has a delay characteristic of the delay time d,
It has flat characteristics (gain 1 characteristics) in the audio frequency band. Reference numeral 9 denotes an arithmetic unit which outputs an output signal y (n) of the adaptive signal processing unit 7 from a signal s (n) output from the delay processing unit 8.
Is subtracted to output an error signal e (n).

At the time of learning, only the noise Xn (z) generated by the automobile CR is input to the microphone 1. The transmission circuit 5
A characteristic CS1 ＾ that simulates a transfer characteristic CS1 from the speaker's mouth to the microphone output end is added to random noise as a target signal. The addition unit 6 combines the output of the transmission circuit 5 and the microphone output, and the adaptive signal processing unit 7 uses the output signal of the addition unit 6 as a reference signal x (n) so that the power of the error signal e (n) is minimized. The adaptive signal processing is performed to update the coefficient W of the adaptive filter 7b. The coefficient W obtained at the end of the learning is set as a noise reduction coefficient in the digital filter 21 of FIG. 3, and the output signal of the calculation unit 9 is input to a speech recognition unit (not shown). According to such a noise reduction device, noise is reduced during voice recognition, and a large speaker voice signal is obtained, so that the SN ratio can be improved.

[0008]

However, in the noise reduction device shown in FIG. 3, the noise reduction coefficient selected according to the vehicle may not always be a coefficient showing optimum characteristics. That is, conventionally, there is a problem that an optimum coefficient for a vehicle cannot be selected from a large number of prepared noise reduction coefficients. This generally means that users
This is because they do not have a measuring instrument or the like and do not know what characteristics the noise component of their car has, and thus do not know what noise reduction coefficient to select. Therefore, noise reduction coefficients can be given to all commercially available vehicles in advance, but the data amount becomes enormous. As described above, an object of the present invention is to provide a noise reduction system that selects and sets a noise reduction characteristic that is optimal for a vehicle of interest.

[0009]

SUMMARY OF THE INVENTION The present invention is a noise reduction system for removing a noise signal from a detected sound signal, wherein (1) a sound detector for detecting a sound in a vehicle interior, and (2) an output from the sound detector. A noise reduction unit that applies a predetermined noise reduction characteristic to the detected sound signal to remove the noise signal from the detected sound signal, (3) a coefficient storage unit that stores a plurality of noise reduction coefficients, (4) the noise reduction coefficient A noise data storage unit that stores noise data that specifies a noise signal referred to when determining each of the noise data in association with the coefficient, (5) noise data storage that is most similar to the noise signal detected by the sound detection unit. Similar noise determination unit for finding the noise signal stored in the unit, (6)
Means for reading a noise reduction coefficient corresponding to the obtained noise signal from the coefficient storage unit and inputting the read noise reduction coefficient to the noise reduction unit, wherein the noise reduction unit applies a predetermined noise reduction characteristic to the detected sound signal based on the input noise reduction coefficient. To remove a noise signal from the detected sound signal.

According to the present invention, a plurality of noise reduction coefficients are stored, and noise data for specifying a noise signal referred to when determining the noise reduction coefficients is stored in association with the coefficients. The noise signal stored in the noise data storage unit most similar to the noise signal detected by the noise signal storage unit is obtained, a noise reduction coefficient corresponding to the noise signal is read from the coefficient storage unit, and input to the noise reduction unit. Is configured to apply a predetermined noise reduction characteristic to the detected sound signal based on the input noise reduction coefficient so as to remove the noise signal from the detected sound signal, so that the noise is removed using the optimal noise reduction coefficient. Thus, the speech recognition rate can be improved.

[0011]

DESCRIPTION OF THE PREFERRED EMBODIMENTS (a) Overall Configuration FIG. 1 is a configuration diagram of a noise reduction system according to the present invention. Reference numeral 51 denotes a microphone for voice detection provided in a vehicle cabin, in addition to a voice emitted by a driver. , Noise in the vehicle compartment (running noise, engine noise, etc.) is detected. Reference numeral 52 denotes a noise reduction unit that removes a noise signal from the detected sound signal, and is a FIR digital filter whose coefficient can be set. 53 is a voice recognition device, 54 is a coefficient bank for storing a plurality of noise reduction coefficients, and 55 is a noise data bank for storing a plurality of noise data. Noise reduction characteristics for removing noise are classified into several types according to the displacement of the automobile and the like.
An adaptive filter coefficient at the time of coefficient convergence, that is, a noise reduction coefficient is obtained for a predetermined vehicle belonging to each class by using the noise reduction coefficient / noise calculation system shown in FIG. Further, noise data for specifying the noise signal referred to when determining the noise reduction coefficient is stored in the noise data bank 55 in association with the noise signal.

FIG. 2 is a block diagram of a noise reduction coefficient / noise calculation system, and an adaptive filter coefficient calculation unit 6
1. A noise data generator 62 is provided. The adaptive filter coefficient calculation unit 61 has the same configuration as the noise reduction system using adaptive signal processing in FIG. The noise data generation unit 62 includes a window function unit 62a that applies a window function to the noise signal detected by the microphone 51, and an FFT calculation unit 62b that performs N-point FFT processing on the noise signal and converts the noise signal into a frequency component (spectral signal). , And a power calculation unit 62c that calculates a power spectrum from each of the spectrum signals at the N points. The power spectrum is expressed in a vector format having N / 2 elements.

The noise reduction coefficient / noise calculation system shown in FIG. 2 is mounted on a predetermined vehicle belonging to each class, and when the vehicle is running at a predetermined speed at a constant speed, the tap length N at which the error signal e (n) is minimized. A filter coefficient W _CNR is obtained, and the filter coefficient W _CNR is output as a noise reduction coefficient. Further, the noise signal at that time is subjected to N-point FFT processing to be converted into a frequency component (spectral signal), and the power spectrum of the N / 2-point noise signal from each of the N-point spectral signals is output as noise data. These noise reduction coefficients W _CNR and noise data are shown in FIG.
Are registered in the coefficient bank 54 and the noise data bank 55, respectively.

Returning to FIG. 1, reference numeral 56 denotes a noise data generator for calculating and outputting the power of the noise signal. The noise data generator 56 has the same configuration as the noise data generator 62 shown in FIG. The power spectrum of the noise signal is output as noise data. A pattern matching unit 57 compares and matches the detected noise data with each of the noise data patterns stored in the noise data bank 55, and outputs a similarity. A determination unit 58 obtains noise data stored in the noise data bank 55 having the highest similarity and outputs an identification number of the noise data.
Reference numeral 9 denotes a coefficient selection unit for selecting a noise reduction coefficient corresponding to the identification number of the noise data from the coefficient bank 54;
Is a noise reduction characteristic determining unit that stores the selected coefficient. The noise reduction unit 52 operates with reference to the noise reduction coefficient written in the characteristic determination unit 60 and removes a noise signal component included in the detected sound signal. Note that the processing of the noise reduction unit 52 may be processing in the time domain, processing in the frequency domain, or processing in the power domain. A noise reduction coefficient is stored in the coefficient bank 54. The coefficient has a data format corresponding to the noise reduction coefficient depending on the configuration of the noise reduction unit.

(B) Overall Operation The coefficient bank 54 and the noise data bank 55 are stored in advance in FIG.
A plurality of noise reduction coefficients and noise data are registered using the system of (1). In this state, the pattern matching unit 57 compares and matches the pattern of the noise data calculated by the noise data generating unit 56 with the pattern of each noise data stored in the noise data bank 55, and outputs the similarity. In this case, the similarity is calculated as follows. When the noise spectrum is given by a vector at N points, the power spectrum Pmic of the noise signal input from the microphone 51 is represented by Pmic = [a ₀ , a ₁ , a ₂ ,..., A _N−1 ]. 1), and the power spectrum Pnoise of the noise data registered in the noise data bank 55 is Pnoise = [b ₀ , b ₁ , b ₂ ,..., B _N−1 ] (2). Further, the weighting coefficient C is set as C = [c ₀ , c ₁ , c ₂ ,..., C _N−1 ] (3).

The value of the weighting coefficient C is set so that the important band is weighted by the speech recognition device. For example,
Assuming that a frequency band of 300 Hz or more is set to be weighted, Pmic and Pnoise are equal to the sampling frequency fs.
= 11.025 kHz, if the power spectrum information of the 128 points, weighted after c _7. Also, define the variable w and (1)
~ (3) evaluation value using the equation defined by following equation _{(similarity) e e = Σ n c n} · (a n -wb n) 2 (n = 0~N-1) (4). From equation (4), the more similar the level and pattern of the power spectrum Pmic and the power spectrum Pnoise are, the smaller the similarity e becomes, and the variable w becomes closer to 1. Also, the power spectrum Pmic and the power spectrum Pn
Even if the oise levels are different, if the patterns are similar, the similarity e will be smaller, but the variable w will be more than one.

As described above, the pattern matching unit 57 obtains the value of the variable w that minimizes the similarity e in the equation (4) and the similarity emin at that time, and sends them to the determination unit 58. Noise data Pnoise
Exist as many as the number stored in the noise data bank 55, so that the pattern matching unit 57
Calculate emin, w for ise and send them to the determination unit 58. The determination unit 58 determines the minimum noise data among all emin, and sends the result (the identification number of the noise data) to the coefficient selection unit 59. In this case, if two or more have the same minimum emin, the variable w is selected closer to 1 and sent to the coefficient selector 59.

The coefficient selection section 59 selects a noise reduction coefficient corresponding to the identification number of the noise data, and sends it to the noise reduction characteristic determination section 60. The noise reduction characteristic determination unit 60 stores the received noise reduction coefficient, and the noise reduction unit 52 applies the noise reduction characteristic corresponding to the coefficient stored in the noise reduction characteristic determination unit 60 to the detected sound signal, A noise signal is removed from the signal and the signal is input to the speech recognition device 53. As described above, the present invention has been described with reference to the embodiments. However, the present invention can be variously modified in accordance with the gist of the present invention described in claims, and the present invention does not exclude these.

[0019]

As described above, according to the present invention, a plurality of noise reduction coefficients are stored, and noise data for specifying a noise signal referred to when determining the noise reduction coefficients is stored in association with the coefficients. Find the noise signal stored in the noise data storage unit most similar to the noise signal detected by the sound detection unit, read the noise reduction coefficient corresponding to the noise signal from the coefficient storage unit and input it to the noise reduction unit, The noise reduction unit applies predetermined noise reduction characteristics to the detected sound signal based on the input noise reduction coefficient so as to remove the noise signal from the detected sound signal, so that the noise is reduced by using the optimum noise reduction coefficient. Can be removed to improve the speech recognition rate.

[Brief description of the drawings]

FIG. 1 is a configuration diagram of a noise reduction system of the present invention.

FIG. 2 is a configuration diagram of a noise reduction coefficient / noise calculation system.

FIG. 3 is a configuration diagram of a conventional noise reduction device.

FIG. 4 is a configuration diagram of a system for obtaining noise reduction characteristics of a conventional noise reduction device.

[Explanation of symbols]

 51 Microphone for voice detection 52 Noise reduction unit 53 Voice recognition device 54 Coefficient bank 55 Noise data bank 56 Noise data generation unit 57 Pattern matching unit 58 Judgment unit 59 ..Coefficient selector 60.Noise reduction characteristic determiner

Claims (1)

[Claims] 1. A noise reduction system for removing a noise signal from a detected sound signal, comprising: a sound detector for detecting a sound in a vehicle interior; and a predetermined noise reduction characteristic applied to a detected sound signal output from the sound detector. A noise reduction unit that removes a noise signal from the detected sound signal; a coefficient storage unit that stores a plurality of noise reduction coefficients; noise data that specifies a noise signal referred to when determining each of the noise reduction coefficients corresponds to the coefficient A noise data storage unit for storing and storing the noise signal; a similar noise determination unit for obtaining a noise signal stored in the noise data storage unit most similar to the noise signal detected by the sound detection unit; and a noise corresponding to the obtained noise signal. Means for reading the reduction coefficient from the coefficient storage unit and inputting the same to the noise reduction unit; Wherein the noise reduction unit applies a predetermined noise reduction characteristic to the detected sound signal based on the input noise reduction coefficient to remove the noise signal from the detected sound signal.